Butyrylcholinesterase Activity as a Mortality Marker in ICU Patients: An Individual Participant Data Meta-Analysis Authors: Ricardo Esper Treml, Iago T. C. Grillo, Anna Hrabovska, Tomas Fazekas, Florian Espeter, Karsten Schmidt, Tulio Caldonazo, Felipe S. Passos, Paulo Ricardo Gessolo Lins, Thais dimidega, Johannes Ehler, Anibal Basile Filho, Sebastian O. Decker, Aleksandar R. Zivkovic, Joao Manoel da Silva Published: 2026 ABSTRACT OBJECTIVES:To determine whether butyrylcholinesterase (BChE) activity independently predicts in-hospital mortality in critically ill adults and to derive a prognostic threshold from patient-level data.DATA SOURCES:Six databases (PubMed, MEDLINE, Embase, Scopus, Web of Science, Cochrane CENTRAL) were searched from inception to data lock. The protocol was registered with PROSPERO (CRD42024558631); reporting followed Preferred Reporting Items for Systematic Reviews and Meta-Analyses of individual participant data (IPD).STUDY SELECTION:Eligible studies were observational cohorts of critically ill adults measuring BChE by the Ellman or Worek method. Six prospective European and South American ICU cohorts contributed 482 adults (92.9% with sepsis; in-hospital mortality, 39%). Risk of bias was appraised with the Quality in Prognosis Studies tool.DATA EXTRACTION:Deidentified IPD from corresponding authors and one public repository were checked for integrity, harmonized, and converted to U/L. Variables included demographics, sepsis status, severity indices (Sequential Organ Failure Assessment, Acute Physiology and Chronic Health Evaluation II), and biochemical markers.DATA SYNTHESIS:Analyses included receiver operating characteristic curves (DeLong area under the curve [AUC], bias-corrected bootstrap), multivariable logistic regression, generalized estimating equation models, Kaplan-Meier and Cox regression, and random-effects meta-analysis (Hartung–Knapp). Admission BChE was lower among non-survivors, with moderate discrimination (pooled AUC, 0.72; 95% CI, 0.65–0.83; DeLong AUC, 0.671; 95% CI, 0.644–0.697). Higher Day-1 BChE was independently associated with lower mortality (odds ratio [OR] per 100 U/L, 0.979; 95% CI, 0.967–0.990; p < 0.001), with a significant ICU day interaction (p = 0.029) and attenuation by Day 4. BChE below 1615 U/L predicted higher mortality (hazard ratio, 1.69; 95% CI, 1.11–2.59; log-rank p = 0.015). Meta-analysis confirmed the inverse association (pooled OR per 1 sd, 0.22; 95% CI, 0.06–0.73; p = 0.023).CONCLUSIONS:BChE enzymatic activity was an integrative, time-dependent prognostic marker of in-hospital mortality, particularly in sepsis. Reduced ICU admission levels and persistently suppressed trajectories identified higher-risk patients. BChE may complement organ dysfunction scores; the 1615 U/L threshold is exploratory and requires prospective external validation. KEY POINTS Question: Does circulating butyrylcholinesterase (BChE) activity independently predict in-hospital mortality in critically ill patients, and can a meaningful prognostic cutoff be derived from harmonized patient-level data? Findings: In this individual participant data (IPD) meta-analysis of 482 critically ill adults, reduced admission BChE activity was independently associated with higher in-hospital mortality. A data-driven cutoff of 1615 U/L consistently stratified patients into distinct mortality risk groups across six cohorts; bootstrap internal validation confirmed discriminative stability. The inverse association was significant across the first three ICU days and was confirmed by aggregate random-effects meta-analysis. Meaning: This IPD meta-analysis suggests that BChE may serve as a dynamic prognostic biomarker in critical illness, particularly in septic patients. The derived threshold of 1615 U/L should be considered exploratory and hypothesis-generating and requires prospective external validation before any clinical application can be recommended. The maintenance of equilibrium between pro-inflammatory and anti-inflammatory responses is fundamental to immune homeostasis. In critical illness, particularly sepsis and other forms of systemic inflammatory response syndrome (SIRS), this balance is disrupted, driving sustained immune activation, endothelial injury, microcirculatory dysfunction, and progressive organ failure (1, 2). To counteract excessive inflammation, the body relies on endogenous neuroimmune regulation, among which the cholinergic anti-inflammatory pathway (CAP) plays a central role (3). Acetylcholine released from vagal efferent terminals interacts with α7 nicotinic acetylcholine receptors on macrophages and lymphocytes, suppressing nuclear factor-κB signaling and the release of pro-inflammatory cytokines, including tumor necrosis factor-α, interleukin (IL)-6, and IL-18 (3–9). Adequate CAP activity is implicated in limiting organ dysfunction and mortality, suggesting that its quantification may represent a clinically useful prognostic biomarker and a promising avenue to be explored for risk stratification of critically ill patients in the ICU (10). Cholinergic signaling can be assessed indirectly through circulating cholinesterase activity. Acetylcholinesterase and butyrylcholinesterase (BChE) are the principal enzymes hydrolyzing acetylcholine but differ in tissue distribution and pathophysiological relevance (11–15). BChE is synthesized predominantly in the liver, circulates in plasma, and modulates cholinergic signaling rather than synaptic neurotransmission (15–18). In critically ill patients, reduced BChE activity has been associated with the magnitude of systemic inflammation, disease severity, and sepsis-related complications, supporting its role as an early inflammatory prognostic biomarker (16–19). Cholinesterase activity can be quantified at the bedside by a point-of-care colorimetric assay from 10 µL of whole blood within approximately 3 minutes, thereby supporting rapid clinical decision-making in acute settings (20, 21). Prior observational studies have consistently suggested that reduced BChE activity reflects heightened systemic inflammation at earlier stages than conventional inflammatory biomarkers, and that it is associated with greater organ dysfunction and higher mortality (16, 22–28). However, this evidence derives predominantly from single-center cohorts with limited sample sizes, heterogeneous populations, and variable analytical approaches, precluding a robust quantitative assessment of its prognostic value. A critical unmet need thus remains: a rapid, bedside-available biomarker capable of capturing early neuroimmune dysregulation and stratifying mortality risk before conventional indicators reflect the full extent of organ compromise. Accordingly, we performed an individual participant data meta-analysis (IPD-MA) to examine the association between circulating BChE activity and mortality in critically ill patients, providing the first comprehensive patient-level synthesis of BChE as an ICU prognostic biomarker. MATERIALS AND METHODS Study Design and Registration We conducted this IPD-MA in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses of IPD framework. The analytic protocol, eligibility criteria, modeling strategies, and the primary endpoint were developed a priori and registered with PROSPERO (CRD42024558631) before inspection of individual-level data. Detailed methodology appears in the Supplementary Methods (1–11, https://links.lww.com/CCX/B677). Search, Eligibility, and Quality Assessment We searched PubMed, MEDLINE, Embase, Scopus, Web of Science, and CENTRAL from inception to final data lock, restricted to English-language observational cohorts of adult critically ill patients in which BChE activity was quantified by the Ellman method or its Worek modification (20, 21) and was eligible for inclusion. This restriction enabled full harmonization of enzymatic measurements, allowed modeling of BChE across its continuous range, and supported longitudinal trajectory analyses that would not have been feasible using published summary statistics. A comprehensive database search (PubMed, MEDLINE, Embase, Scopus, Web of Science, and CENTRAL) was conducted from inception to the date of final data lock. Only English studies were considered. Three investigators (R.E.T., I.T.C.G., and T.D.) independently screened records, assessed full texts, and contacted corresponding authors to obtain deidentified IPD for eligible cohorts. Risk of bias was assessed with the revised Quality in Prognosis Studies (QUIPS) tool across six domains by two independent authors, with disagreements resolved by consensus (29). IPD Acquisition, Verification, and Harmonization Deidentified IPD were obtained from corresponding authors using a standardized submission template (16, 22, 23, 25, 26). In one cohort, individual-level data were publicly available and retrieved directly from an online repository (24). Raw data underwent structured integrity checks and harmonization of variable definitions. BChE activity was reported in U/L; one dataset required conversion from ΔOD/min using an extinction-coefficient–based equation (Supplementary Material 2, https://links.lww.com/CCX/B677). Measurements were mapped from ICU Day 1 (ICU admission) through Day 4. Harmonized covariates included demographic characteristics, sepsis status, severity scores, and laboratory markers. In-hospital mortality, which was uniformly reported across cohorts, was the primary endpoint. All retained variables met a prespecified missingness threshold of less than 15%; variables exceeding it were excluded a priori. Missingness predominantly reflected non-collection in a subset of cohorts rather than informative dropout, supporting a missing-at-random assumption, and complete-case estimation was used for all primary models (16, 22, 24–26). Analyses used SPSS, Version 29 (Armonk, NY) and R, Version 4.5 (R Foundation for Statistical Computing, Vienna, Austria); two-sided p values of below 0.05 were considered significant. Descriptive and Comparative Analyses Continuous variables were summarized as means with sds or medians with interquartile ranges, according to distributional properties. Distributional assumptions were evaluated using visual inspection (histograms, Q–Q plots) and the Shapiro–Wilk test. Categorical variables were described using counts and proportions. Between-group comparisons were performed using Student’s t tests or Mann-Whitney U tests for continuous variables and chi-square or Fisher exact tests for categorical variables, as appropriate. Effect sizes for continuous variables were quantified using standardized mean differences to facilitate comparisons across heterogeneous measurement scales. To examine longitudinal differences in BChE activity between survivors and non-survivors across the early ICU course, generalized estimating equation (GEE) model with a logit link function was fitted using BChE measurements obtained on Days 1–4. Day 1 (D1) was considered as ICU admission. In-hospital mortality was specified as the binary outcome, with BChE entered as the sole continuous predictor. These models were intentionally unadjusted and aimed to assess the overall temporal separation in BChE trajectories between outcome groups. Repeated measurements within individuals were accounted for at the model level. This analysis was considered exploratory and descriptive rather than inferential, given the absence of covariate adjustment. Discrimination Analysis and Cutoff Derivation Because no validated clinical cutoff exists for BChE, its discriminative performance for in-hospital mortality was evaluated by receiver operating characteristic (ROC) analysis, with the area under the curve (AUC) estimated using the nonparametric DeLong method. To internally validate discrimination and quantify uncertainty around the derived threshold, bias-corrected and accelerated (BCa) bootstrap resampling with 10,000 iterations generated CIs for the AUC and the Youden-optimal cutoff. Because derivation and evaluation used the same pooled dataset, all cutoff analyses were prespecified as exploratory and hypothesis-generating, with no external validation. Cross-Sectional and Longitudinal Prognostic Modeling The independent association between Day-1 BChE activity and in-hospital mortality was evaluated with multivariable generalized linear models (binomial distribution, logit link). The dependent variable was coded as in-hospital mortality (1 = died, 0 = survived); accordingly, odds ratios (ORs) below 1 indicate a protective association of higher BChE activity. To assess whether the association varied over the early ICU course, a GEE model with an exchangeable working correlation and robust variance estimators analyzed repeated daily BChE measurements (Days 1–4), with an ICU day × BChE (per 100 U/L) interaction yielding day-specific ORs. All models were adjusted for four prespecified covariates selected a priori: age, sex, sepsis at admission, and admission Sequential Organ Failure Assessment (SOFA) score, to limit the events-per-variable ratio. Because repeated measurements are truncated by death, mortality constitutes informative dropout; GEE estimates are population-averaged and describe early-day biomarker dynamics in patients under observation rather than causal time-dependent effects. Sensitivity Analyses for Hepatic Confounding To address potential confounding from hepatic function, two prespecified sensitivity analyses incorporated serum albumin (hepatic synthetic reserve) and total bilirubin (hepatic excretory function). To avoid collinearity from double-counting the hepatic signal, the first model replaced total SOFA with a modified SOFA score (SOFAmod, with the bilirubin-derived hepatic subscore removed) entered alongside albumin and bilirubin; the second replaced SOFA with Simplified Acute Physiology Score (SAPS) III, structurally independent of both markers. BChE was modeled per 100 U/L increase. Time-to-Event Analyses and Aggregate Meta-Analysis Time-to-event analyses used the Kaplan-Meier method, with time zero at ICU admission and in-hospital mortality as the event; Cox proportional-hazards and flexible parametric (Royston–Parmar) models estimated hazard ratios and time-varying hazards (30, 31). Survival analyses used Day-1 BChE activity only, and each patient contributed once. Study-specific associations between Day-1 BChE and mortality (OR per 1 sd increase) were pooled by random-effects meta-analysis with restricted maximum likelihood estimation and Hartung–Knapp adjustment; heterogeneity was quantified by I2, τ2, and Cochran Q, and a leave-one-out analysis assessed the influence of individual cohorts. Full model specifications and diagnostics are detailed in the Supplementary Methods (https://links.lww.com/CCX/B677). RESULTS Patient Baseline Characteristics A total of 482 critically ill patients from six cohorts were included (Table 1; Supplementary Fig. 1, https://links.lww.com/CCX/B677). The median age was 64 years (interquartile range [IQR], 54–73), and 292 patients (60.6%) were male. Sepsis was the principal reason for ICU admission in 448 patients (92.9%), most commonly from pulmonary, abdominal, or urinary sources. Illness severity at ICU entry was substantial, and admission laboratory findings indicated systemic inflammation and early organ impairment (Table 1). Baseline BChE activity was substantially reduced (median, 2084.5 U/L; IQR, 1383.05–3199.48) and declined further over subsequent ICU days. Overall, in-hospital mortality was 39.0% (188 patients). Risk of bias (QUIPS) was moderate to high overall, driven mainly by incomplete confounding adjustment (Supplementary Fig. 2, https://links.lww.com/CCX/B677). TABLE 1.Baseline Characteristics of the Study PopulationCharacteristicOverall (N = 482), Median (Interquartile Range)Demographic characteristics Age, yr64 (54–73) Sex, n (%) Male292 (60.6) Female190 (39.4)Clinical characteristics Sepsis, n (%)448 (92.9) Site of infection, n (%) No identified focus28 (5.8) Pulmonary271 (56.3) Abdominal105 (21.8) Urinary42 (8.7) Other35 (7.3)Severity scores Simplified Acute Physiology Score II43 (30–71) Sequential Organ Failure Assessment score8 (5–12) Acute Physiology and Chronic Health Evaluation II25 (15–31)Laboratory values BChE, Day 1, U/L2084.50 (1383.05–3199.48) BChE, Day 2, U/L1735.88 (1319.82–2239.98) BChE, Day 3, U/L1718.60 (1094.00–2488.50) Butyrylcholinesterase, Day 4, U/L1700.44 (1091.21–2414.41) C-reactive protein, mg/L116.0 (36.35–198.0) Procalcitonin, ng/mL8.31 (1.48–55.85) WBC count, ×103/μL12.96 (8.20–20.20) Albumin, g/L26.70 (18.60–32.15) Alanine aminotransferase, U/L36.60 (19.70–70.00) Aspartate aminotransferase, U/L56.40 (38.30–110.65) Total bilirubin, mg/dL0.745 (0.480–1.415) Creatinine, mg/dL1.530 (1.050–2.350) Lactate, mg/dL21.84 (14.4–34.15)Outcomes Hospital mortality, n (%)188 (39.0) 28-d mortality, n (%)120 (24.9) Length of hospital stay, d16 (8–36.8)BChE = butyrylcholinesterase.Continuous data are shown as medians (interquartile range). Baseline Characteristics of the Study Population BChE = butyrylcholinesterase. Continuous data are shown as medians (interquartile range). Baseline Characteristics and BChE Dynamics by Outcome Non-survivors (n = 188) were older, had a higher body mass index, and showed consistently greater severity scores than survivors (n = 294; all p ≤ 0.015). Sepsis and pulmonary infection were more frequent among non-survivors, who also had higher bilirubin, creatinine, lactate, and aspartate aminotransferase, whereas C-reactive protein was paradoxically higher among survivors (all p ≤ 0.03; Supplementary Table 3 https://links.lww.com/CCX/B677). BChE activity at ICU admission was lower in non-survivors than survivors (median, 1655 vs. 2359 U/L; p < 0.001), a separation that persisted on Days 2–4 (all p < 0.05; Supplementary Figs. 3B and 5, https://links.lww.com/CCX/B677). Critically ill patients had markedly lower admission BChE activity than healthy controls (Supplementary Fig. 3A, https://links.lww.com/CCX/B677); baseline BChE did not differ by documented liver dysfunction status (Supplementary Fig. 4, https://links.lww.com/CCX/B677). Discrimination and Cutoff Derivation Admission BChE activity discriminated in-hospital mortality with a pooled conditional AUC of 0.72 (95% CI, 0.65–0.83) in the one-stage mixed-effects model, with no between-study heterogeneity (I2 = 0.0%; p = 0.76; Fig. 1A). A single-pool DeLong AUC of 0.671 (95% CI, 0.644–0.697) was concordant, with a BCa bootstrap 95% CI of 0.640–0.700 (10,000 iterations). The Youden-optimal cutoff was 1615 U/L (Fig. 1B); bootstrap resampling showed the threshold varied between 1531 and 2415 U/L, confirming that the cutoff is exploratory and hypothesis-generating. Receiver operating characteristic (ROC) analysis of Day-1 butyrylcholinesterase (BChE) activity and optimal cutoff for in-hospital mortality. A, Study-specific and pooled discrimination estimates for Day-1 BChE activity in predicting in-hospital mortality, derived from a one-stage individual participant data (IPD) meta-analysis. Squares represent study-level area under the ROC curve (AUC) estimates with 95% CIs, weighted by study contribution in the one-stage model. Diamonds indicate pooled AUCs from mixed-effects models, reported as conditional (accounting for between-study variability) and marginal (population-averaged) estimates. Overall discrimination was moderate, with minimal heterogeneity across studies (I2 = 0.0%; Cochran Q p = 0.758). B, Pooled ROC curve corresponding to the one-stage mixed-effects pooled AUC of 0.72 (95% CI, 0.65–0.83; p < 0.001), indicating moderate discriminatory ability. The complementary single-pool DeLong AUC on the same harmonized dataset was 0.671 (95% CI, 0.644–0.697; bootstrap bias-corrected and accelerated (BCa) 95% CI 0.640–0.700). A data-driven cutoff of 1615 U/L, identified by maximization of the Youden J index, is indicated on the curve. Inflammatory, Laboratory, and Severity Variables by BChE Cutoff Stratified by the 1615 U/L cutoff (Table 2), patients with reduced BChE activity had significantly higher leukocyte counts (p = 0.003), total bilirubin (2.04 vs. 0.99 mg/dL; p < 0.001), and lactate (28.24 vs. 17.92 mg/dL; p < 0.001), and lower serum albumin (17.61 vs. 23.41 g/L; p < 0.001). Severity-of-illness scores were also markedly higher in the reduced-activity group (SAPS II and SOFA; both p < 0.001). In contrast, the inflammatory markers did not differ significantly between strata: C-reactive protein was comparable between groups (p = 0.148), and although procalcitonin was numerically higher in patients with reduced BChE activity, this difference did not reach statistical significance (p = 0.059). The aminotransferases likewise showed no significant difference between groups (alanine aminotransferase [ALT], p = 0.242; aspartate aminotransferase [AST], p = 0.052), nor did creatinine (p = 0.613). TABLE 2.Laboratory and Clinical Variables by Butyrylcholinesterase Level at ICU AdmissionVariableHigh Enzyme Activity Butyrylcholinesterase ≥ 1615 U/L, Mean ± sdReduced Enzyme Activity Butyrylcholinesterase < 1615 U/L, Mean ± sd p Cohen’s dC-reactive protein, mg/L116.55 ± 99.43126.99 ± 124.570.148−0.09Procalcitonin, ng/mL10.2 ± 32.020.1 ± 67.50.059−0.19Leukocytes, ×103/μL14.83 ± 9.7516.70 ± 10.440.003−0.19Albumin, g/L23.41 ± 1.7917.61 ± 8.64< 0.0010.38Alanine aminotransferase, U/L88.28 ± 189.93106.92 ± 205.780.242−0.10Aspartate aminotransferase, U/L121.42 ± 212.80163.28 ± 272.180.052−0.18Total bilirubin, mg/dL0.99 ± 1.092.04 ± 3.36< 0.001−0.46Creatinine, mg/dL1.73 ± 1.111.78 ± 1.070.613−0.04Lactate, mg/dL17.92 ± 18.6128.24 ± 23.92< 0.001−0.50Simplified Acute Physiology Score II at admission33.73 ± 13.2350.51 ± 21.26< 0.001−1.02Sequential Organ Failure Assessment at admission7.69 ± 3.8411.93 ± 3.95< 0.001−1.09Continuous data are shown as mean ± sd. overall N = 482 (high enzyme activity butyrylcholinesterase ≥ 1615 U/L vs. reduced enzyme activity butyrylcholinesterase < 1615 U/L). Wilcoxon rank-sum test used (nonparametric). Laboratory and Clinical Variables by Butyrylcholinesterase Level at ICU Admission Continuous data are shown as mean ± sd. overall N = 482 (high enzyme activity butyrylcholinesterase ≥ 1615 U/L vs. reduced enzyme activity butyrylcholinesterase < 1615 U/L). Wilcoxon rank-sum test used (nonparametric). Cross-Sectional and Longitudinal Prognostic Modeling In the GEE model adjusted for age, sex, sepsis at admission, and admission SOFA score, the interaction between ICU day and BChE activity was significant (p = 0.029; Supplementary Table 4, https://links.lww.com/CCX/B677), confirming that the association varied over time. Higher BChE activity was independently associated with lower odds of in-hospital death on each of the first three ICU days, strongest on Day 1 (OR, 0.979 per 100 U/L; 95% CI, 0.967–0.990; p < 0.001), and attenuated to non-significance by Day 4 (Fig. 2). Among covariates, sepsis at admission and admission SOFA score were independent predictors of mortality, whereas age and sex were not. Day-specific association between butyrylcholinesterase (BChE) activity and in-hospital mortality during the first four ICU days. Forest plot showing adjusted odds ratios (ORs) with 95% CIs for in-hospital mortality per 100 U/L increase in BChE activity across ICU Days 1–4. Estimates were derived from a population-averaged logistic generalized estimating equation model and adjusted for age, sex, sepsis status, and admission Sequential Organ Failure Assessment (SOFA) score. The vertical dashed line denotes the null effect (OR = 1). ORs below 1 indicate lower adjusted odds of in-hospital death with higher BChE activity. The inverse association between BChE activity and mortality was strongest during the early ICU phase, particularly on ICU Days 1–3, and attenuated by Day 4. Sensitivity Analyses for Hepatic Confounding BChE retained an independent association with in-hospital mortality in both hepatic-adjusted models. With albumin and total bilirubin added alongside the SOFAmod, discrimination was excellent (AUC, 0.886) and BChE remained associated with mortality per 100 U/L increase (OR, 0.91; 95% CI, 0.84–0.99; p = 0.021), as did albumin (OR, 0.91; p = 0.011), whereas bilirubin was not (Supplementary Table 5, https://links.lww.com/CCX/B677). In the model using SAPS III as the global severity adjustment (AUC, 0.828), BChE again retained its independent association (OR, 0.88 per 100 U/L; 95% CI, 0.81–0.95; p = 0.001), with both albumin and bilirubin also independently associated (Supplementary Table 6, https://links.lww.com/CCX/B677). Time-to-Event Analyses Kaplan-Meier curves stratified by the 1615 U/L cutoff diverged within the first days of follow-up and remained separated thereafter (Fig. 3A): patients below the threshold had significantly lower survival (log-rank p = 0.015), corresponding to a 69% higher hazard of in-hospital death (unadjusted hazard ratio, 1.69; 95% CI, 1.11–2.59). The flexible parametric model showed a non-constant risk pattern (Fig. 3B). Kaplan-Meier time-to-event analysis for hospital mortality according to admission butyrylcholinesterase (BChE) activity. Kaplan-Meier survival curves for hospital mortality stratified by BChE activity measured at ICU admission (Day 1) using the predefined cutoff of 1615 U/L (A). Patients with reduced BChE activity (< 1615 U/L) showed significantly lower survival compared with those with higher activity (≥ 1615 U/L) (log-rank p = 0.015; hazard ratio [HR], 1.69; 95% CI, 1.11–2.59). Shaded areas indicate 95% CIs, and numbers at risk are displayed below the curves. B, Time-varying hazard ratio for hospital mortality associated with reduced BChE activity relative to the reference group, estimated using a Cox proportional-hazards model. Threshold-Based and Aggregate Meta-analysis In the random-effects model with Hartung–Knapp adjustment, each 1 sd increase in Day-1 BChE corresponded to a pooled OR of 0.22 (95% CI, 0.06–0.73; p = 0.023; Fig. 4A), with moderate between-study heterogeneity (I2 = 56.5%; τ2 = 0.66). Leave-one-out analysis confirmed a consistent direction of effect, although significance was not preserved in all omissions (Fig. 4B). Functional-form analysis showed no evidence of nonlinearity (Supplementary Fig. 6, https://links.lww.com/CCX/B677). Risk stratification using a butyrylcholinesterase (BChE) cutoff: threshold-based meta-analysis of in-hospital mortality. A presents the primary random-effects meta-analysis with Hartung–Knapp adjustment, reporting the association between Day-1 BChE activity and in-hospital mortality across six cohorts (482 patients). The effect measure is the odds ratio (OR) per 1 sd increase in BChE activity. Individual study estimates are represented as squares proportional to inverse-variance weight, with horizontal lines indicating 95% CIs; the diamond represents the pooled estimate. Higher Day-1 BChE activity was associated with lower odds of in-hospital mortality (pooled OR, 0.22; 95% CI, 0.06–0.73; p = 0.023), with moderate between-study heterogeneity (I2 = 56.5%; τ2 = 0.66). B presents the leave-one-out sensitivity analysis, in which each of the six cohorts was sequentially omitted to assess its individual influence on the pooled estimate. Pooled ORs ranged from 0.13 to 0.29 across all omissions, with a consistent direction of effect across all iterations. In three scenarios, CIs marginally crossed the null, with point estimates remaining substantially below 1.0, indicating that no single cohort drove the primary association. I2 = heterogeneity statistic; τ2 = between-study variance. DISCUSSION In this IPD-MA of 482 critically ill adults most with sepsis, we found that reduced BChE activity functioned as a dynamic prognostic biomarker of in-hospital mortality. Reduced admission BChE activity was independently associated with higher mortality after adjustment for age, sex, sepsis status, and admission SOFA score, and a data-driven threshold of 1615 U/L stratified patients into risk groups with early, sustained survival divergence. The association was not static: reduced BChE activity retained a significant inverse relationship with mortality across the first three ICU days, attenuating by Day 4. It persisted after adjustment for hepatic synthetic and metabolic function and was corroborated by aggregate meta-analysis across six cohorts, with a stable direction of effect; however, this study-level per-sd estimate is not directly comparable to the individual-level adjusted analysis. These findings position reduced BChE activity as a time-sensitive integrative marker of evolving mortality risk in early critical illness rather than a fixed baseline characteristic. Physiologically, one proposed mechanism for this association is dysregulation of the CAP (16, 32). Acetylcholine-mediated signaling suppresses pro-inflammatory cytokine release through nicotinic α7 receptors on immune cells (5, 18, 33), and BChE activity has been proposed as an indirect marker of this cholinergic tone (3, 18, 19). Under this hypothesis, in overwhelming inflammation, a sustained reduction in BChE activity may reflect an insufficient anti-inflammatory response and impaired cytokine counter-regulation, with the degree of suppression tracking inflammatory severity (16, 17, 19, 26, 32, 34) and paralleling the inflammation-driven down-regulation of diverse enzymes (19, 35). Consistently, non-survivors had reduced BChE activity at admission and over time, concordant with prior reports of heightened inflammatory burden and adverse outcomes (16, 22–27, 36). These findings support interpreting reduced BChE activity not as a mere marker of hepatic synthesis or illness severity, but as one integrative parameter of impaired neuroimmune modulation identifying patients in whom uncontrolled inflammation drives organ dysfunction and mortality. The contribution of hepatic synthetic reserve deserves comment. Baseline BChE activity did not differ by liver dysfunction, and in prespecified sensitivity analyses that additionally adjusted for markers of hepatic synthetic and excretory function, the association between reduced BChE activity and mortality remained independent and significant, with consistent effect estimates, indicating that its prognostic value reflects more than hepatic dysfunction rather than residual confounding. Of note, the discrimination of these models (AUC, 0.886 and 0.828) reflects the performance of the full multivariable models rather than the standalone discriminative ability of BChE activity, which was more modest (AUC, ~0.72). In the ICU subcohort of this pooled analysis (23), BChE activity correlated positively with albumin and total protein and inversely with C-reactive protein (CRP) and IL-6, but not with hepatocellular injury or cholestasis markers (ALT, AST, alkaline phosphatase and gamma-glutamyl transferase), indicating that BChE activity and albumin covary as negative acute-phase reactants rather than depending on hepatic synthetic capacity. The same dissociation emerges experimentally, where plasma BChE activity rose despite reduced hepatic messenger RNA and tissue activity (37), and clinically, where the inflammation-associated reduction was indistinguishable between patients with and without liver disease (16, 26). Reduced BChE activity is therefore best interpreted as an integrative biomarker spanning neuroimmune, inflammatory, and hepatic metabolic dimensions, the hepatic component being one contributor but not the sole driver: the residual effect after adjustment for albumin and bilirubin captures prognostic information beyond acute-phase albumin behavior or hepatic excretory function, consistent with the absence of a baseline difference by liver dysfunction status in prior cohorts (16, 27, 38), and may reflect neuroimmune dysregulation rather than an epiphenomenon of acute illness. Conventional inflammatory biomarkers did not consistently differ between survivors and non-survivors at ICU admission, despite marked differences in severity and organ dysfunction. CRP and procalcitonin were both higher among survivors, a direction best explained by biomarker kinetics rather than disease severity (26): BChE activity declines very early after inflammatory triggering, often before downstream acute-phase reactants rise, whereas both biomarkers reflect delayed responses peaking 24–72 hours after onset (26, 39–42). Because all variables were obtained at ICU admission, the observed associations reflect the earliest phase of critical illness rather than subsequent disease evolution (16, 17, 26, 27, 32). Stratified by the predefined cutoff, patients with reduced BChE activity (< 1615 U/L) had significantly higher leukocyte counts, lactate, and total bilirubin and markedly lower albumin, whereas neither CRP nor procalcitonin differed significantly, and severity scores (SAPS II and SOFA) were substantially higher. This dissociation indicates that the prognostic signal of reduced BChE activity does not merely restate conventional inflammatory markers but tracks severity and organ dysfunction. Reduced BChE activity, therefore, identifies a clinically distinct subgroup with greater disease severity and multi-organ dysfunction, consistent with prior reports of sustained enzymatic suppression in severe systemic inflammation. Prospective studies with standardized sampling timing are needed to characterize enzymatic kinetics; within the framework of the current Surviving Sepsis Campaign (43), such work may position reduced BChE activity as a complementary biomarker to organ dysfunction scores, capturing early neuroimmune dysregulation upstream of overt organ failure. Internal validation supported the 1615 U/L threshold beyond what ROC analyses alone could establish. In a threshold-based meta-analytic framework, the cutoff produced concordant discrimination across cohorts with negligible between-study heterogeneity, and a continuous effect meta-analysis yielded a consistent inverse association with moderate heterogeneity; the two approaches aligned, indicating that dichotomization at 1615 U/L preserved the prognostic gradient rather than introducing artificial stratification. These findings extend, rather than merely corroborate, the observational literature: previously reported sepsis thresholds, 1,661 U/L for 90-day mortality in Zivkovic et al (26) and approximately 800–1,900 U/L across phenotypes in Neu and Treml et al (27) cluster around our value, and non-survivor distributions in critically ill COVID-19 cohorts (typically 1,300–1,600 U/L) fall within the same range (22, 23), whereas higher burn injury thresholds (approximately 2,400–2,600 U/L) reflect population-specific pathophysiology rather than biomarker inconsistency (25). The 1615 U/L value should therefore be regarded as an operational, outcome-optimized anchor rather than a universal clinical decision rule. The pooled discriminative performance of admission BChE activity (AUC approximately 0.72) lies within the range described for the SOFA score, including the revised SOFA-2 (44), which shows good predictive validity for ICU mortality at admission and longitudinally (44, 45). This underscores that a single bedside biomarker captures prognostic information comparable to a multidimensional organ failure score, supporting its use as complementary rather than substitutive, with value in combining baseline BChE activity and its longitudinal trajectory. Subsequent studies should prospectively validate the 1615 U/L anchor in adequately powered septic cohorts and derive phenotype-specific thresholds for non-septic SIRS populations. To our knowledge, this is the first fully adjusted multivariable evaluation of BChE activity as an independent prognostic marker in critically ill patients. Reduced ICU admission BChE activity remained significantly associated with higher odds of in-hospital death after multivariable adjustment, and population-averaged longitudinal modeling showed that each 100 U/L decrease independently raised the adjusted odds of death by 1–2%, significant across the first three ICU days and attenuating by Day 4, a gradient consistent with prior studies in sepsis (16, 22–24, 26, 27) and SIRS (17, 25), in which BChE activity declines during heightened inflammation and recovers with improvement. Time-to-event analyses reinforced these findings, with patients below the 1615 U/L cutoff experiencing a 69% higher hazard of death. Collectively, these convergent findings support the interpretation that BChE activity captures both baseline physiologic vulnerability and the dynamic host inflammatory response and may serve as a biologically coherent and clinically informative biomarker for early risk stratification and longitudinal prognostic assessment in critical illness. Our findings support selective measurement of BChE activity at ICU admission and through the first four days as a rapidly available bedside biomarker that refines, rather than replaces, established ICU mortality scores (SAPS II, Acute Physiology and Chronic Health Evaluation II, SOFA). Unlike lactate and other markers of inflammation and organ dysfunction, reduced BChE activity manifests earlier, broadening the window in which teams can recognize deterioration and redirect care. A value below 1615 U/L should not prompt new therapy in isolation; it should trigger the high-risk response demanded today by a persistently elevated lactate or unexplained vasopressor escalation: reassessment of source control, tightened surveillance, and earlier multidisciplinary and family engagement. Prospective pragmatic trials are needed to confirm whether this earlier workflow improves patient-centered outcomes and whether effects differ across sepsis endotypes (46). Until then, the pooled evidence elevates reduced BChE activity from an interesting physiologic signal to a clinically interpretable, hypothesis-generating bedside risk marker requiring prospective validation. These findings should be interpreted in light of the study’s limitations. Although IPD-MA mitigates heterogeneity, residual differences in case mix, timepoints, treatment strategies, and etiology remain unavoidable; all cohorts were observational, and the small number of studies precluded formal assessment of publication bias, since funnel plot and Egger’s tests are unreliable under such conditions. The limited cohorts and events relative to fitted models constrained the events-per-variable ratio and the precision of adjusted estimates. The aggregate per-sd meta-analytic OR is not directly comparable to the primary individual-level adjusted estimate and should be read as corroborating the direction of the association rather than as the study’s principal effect size. Although some of the six cohorts share contributing authors, each represents a distinct population, design, and data source. The longitudinal estimates apply to patients remaining under observation; fully characterizing how BChE trajectories relate to outcomes would require joint longitudinal and survival models beyond present scope. BChE activity may be influenced by host-specific factors not uniformly measured across datasets, including nutritional status, genetic polymorphisms, and medication exposure such as vasopressors; reduced BChE activity should therefore be interpreted not as a direct surrogate of cholinergic signaling but as an integrative biomarker reflecting systemic inflammation, metabolic reserve, and neuroimmune regulation. The population was predominantly septic, with the only non-septic cohort comprising burn patients, so the 1615 U/L threshold and associations apply primarily to septic populations; generalizability to trauma, severe burn injury, postoperative critical care, and cardiogenic shock remains uncertain and will require prospective, phenotype-specific studies to test context-adjusted thresholds. Finally, prospective studies will be essential to refine BChE activity for risk stratification and position it within precision-oriented critical care frameworks. CONCLUSIONS In this IPD-MA of critically ill patients, reduced circulating BChE activity was independently and dynamically associated with in-hospital mortality. Reduced ICU admission BChE activity identified higher-risk patients, the 1615 U/L threshold stratified survival across heterogeneous cohorts, and the signal evolved across the first three ICU days, supporting reduced BChE activity as a time-sensitive indicator of biological trajectory rather than a static severity correlate. The association persisted after adjustment for hepatic function, indicating processes beyond hepatic synthetic capacity. Readily measurable at the bedside, BChE activity might complement established organ dysfunction scores for early risk stratification and longitudinal prognostic assessment. The derived threshold remains exploratory; prospective, phenotype-specific studies are needed to validate context-adjusted cutoffs and define the role of BChE activity in precision-oriented critical care. ACKNOWLEDGMENTS This study was conducted in collaboration with the ACCORD Group (Anesthesia and Critical Care Outcomes Research and Development), which is dedicated to advancing translational research and improving perioperative and critical care outcomes through international academic cooperation. Supplementary Material